z/OS MVS Initialization and Tuning Guide
Previous topic | Next topic | Contents | Contact z/OS | Library | PDF


Performance recommendations

z/OS MVS Initialization and Tuning Guide
SA23-1379-02

The following recommendations can improve system performance through the careful use of paging data sets and devices.

  1. Allocate only one paging data set per device. Doing this reduces contention among more than one data set for the use of the device. If you do define more than one paging data set for each device, the use of Parallel Access Volume (PAV) devices can help to reduce contention for a device.

    Reason: If there is more than one paging data set on the same device, a significant seek penalty is incurred. Additionally, if the data sets are local page data sets, placing more than one on a device can cause the data sets to be selected less frequently to fulfill write requests.

    Comments: You might, however, place low-activity non-paging data sets on a device that holds a page data set and check for device contention by executing RMF™ to obtain direct access device activity reports during various time intervals. The RMF data on device activity count, percent busy, and average queue length should suggest whether device contention is a problem. RMF data for devices that contain only a page data set can be used as a comparison base.

  2. Over-specify space for all page data sets.

    Reason: Over-specifying space allows for the creation of additional address spaces before the deletion of current ones and permits some reasonable increase in the number of concurrent VIO data sets which may be backed by auxiliary storage. VIO data set growth might become a problem because there is no simple way to limit the total number of VIO data sets used by multiple jobs and TSO/E sessions. VIO data set paging can be controlled by restricting it to certain page data sets through the use of directed VIO.

    VIO data set pages can be purged by a reIPL specifying CVIO (or CLPA). CVIO indicates that the system is to ignore any VIO pages that exist on the page data sets and treat the page data sets initially as if there is no valid data on them (that is, there are no allocated slots). Thus, specifying CVIO prevents the warm start of jobs that use VIO data sets because the VIO pages have been purged. (For additional space considerations, see the guideline for estimating the total size of paging data sets in Estimating total size of paging data sets.)

    In all cases, ASM avoids scattering requests across large, over-specified, page data sets by concentrating its activity to a subset of the space allocated.

  3. Use more than one local page data set, each on a unique device, even if the total required space is containable on one device.

    Reason: When ASM uses more than one data set, it can page concurrently on more than one device. This is especially important during peak loads.

  4. Distribute ASM data sets among channel paths and control units.

    Reason: Although ASM attempts to use more than one data set concurrently, the request remains in the channel subsystem queues if the channel path or control unit is busy.

  5. Dedicate channel paths and control units to paging devices.

    Reason: In heavy paging environments, ASM can use the path to the paging devices exclusively for page-ins and page-outs and avoid interference with other users, such as IMS™.

  6. Make a page data set the only data set on the device.

    Reason: Making a paging data set the only data set on the device enables ASM to avoid contention. ASM can monopolize the device to its best performance advantage by controlling its own I/O processing of that data set. ASM does not have to perform the additional processing that it would otherwise have to perform if I/O for any other data set, especially another page data set, were on the same device. If another data set must be placed on the device, select a low-use data set to minimize contention with the page data set.

  7. Do not share volumes that contain page data sets among multiple systems.

    Reason: While page data sets may be defined on volumes that contain shared non-paging data sets, they cannot be shared between systems.

  8. Take one of the following actions to control the risk of auxiliary storage shortages. Auxiliary storage shortages have severe effects on system performance when they occur, and can also cause the system to fail. When a shortage occurs, the system rejects LOGON, MOUNT, and START commands and keeps address spaces with rapidly increasing auxiliary storage requirements from running until the shortage is relieved.
    1. Use the SMF Step Initiation Exit, IEFUSI, to limit the sizes of most address spaces and data spaces.
    2. If you do not establish limits using IEFUSI, consider over-allocating local page space by an amount sufficient to allow a single address space or data space to reach the virtual storage limit (as might happen if a program looped obtaining storage) without exhausting virtual storage shortage.

    See Example 4: Sizing local page data sets for more information about calculating local page data set space requirements.

Go to the previous page Go to the next page




Copyright IBM Corporation 1990, 2014